Loading…
Several agents, one monorepo
Coordinating several agents on one codebase
What it looks like to run Claude Code, Codex, and Cursor against the same monorepo without them colliding: one workspace, shared task state, and explicit handoffs between agents.
Worked example · mechanism, not a customer story
Table of contents
CHALLENGE
Run more than one coding agent against the same repository and they work in the dark. Each has its own session and its own idea of the current state, so two agents will happily edit the same file from different assumptions. Context does not survive a restart, so every run begins by re-explaining the codebase. The failure is not that any single agent is bad at its job; it is that nothing connects them.
Agents that cannot see each other's work will redo it, or undo it. The missing piece is shared state, not a better model.
SOLUTION
Connect the agent CLIs you already pay for. nala coordinates the tools rather than replacing them, so Claude Code stays Claude Code; it does not resell inference or route your work through a nala-hosted model.
Break the work into tasks that agents claim. A claimed task is visibly owned, which is what stops two agents starting the same thing. Each department works in its own git worktree, so parallel edits do not collide on disk — worktrees are file isolation, not a security boundary.
Hand off through durable agent-to-agent messages. A builder can ask a reviewer to look at a diff, a lead can task its department, and the message survives a restart because it is stored rather than held in a session. Delivery is acknowledged end to end, so a handoff either happened or visibly did not.
Keep the operator in the loop where it matters. An agent cannot spawn another agent without your approval, and each approval is single-use, so a spawn request is a decision rather than a formality.
WHAT YOU GET
The work becomes inspectable: which agent holds which task, what they said to each other, and what changed on disk.
This is a description of mechanism, not a measured result. What it buys you is visibility and the absence of duplicated work; how much time that saves depends entirely on your codebase, your agents, and how you scope tasks. We have no customer measurements to offer, and would rather say so than invent a percentage.
Follow the build
Product updates and new worked examples as they are written. No customer stories until there are customers to quote.
WHAT'S NEXT
Your coding agents are fast. We keep them on track
Brand new and local-first. It extends the subscriptions you already pay for. Be among the first to put a command center behind them.
The Nerve Center for Agentic Engineering